The Economic Determinants of the Cultural Backlash : Globalization and Attitudes in Western Europe

Transcription

1 The Economic Determinants of the Cultural Backlash : Globalization and Attitudes in Western Europe Italo Colantone Piero Stanig August 4, 2018 Abstract We investigate the impact of globalization on people s attitudes in fifteen Western European countries, over We employ data from the European Social Survey (ESS) and the European Values Study (EVS). We compute a time-varying regionspecific measure of exposure to Chinese imports, based on the historical industry specialization of each region. We attribute to each individual the import shock in the region of residence in the years prior to the survey. To identify the causal impact of the import shock, we instrument imports to Europe using Chinese imports to the United States. We find that respondents residing in regions that received stronger globalization shocks are systematically less supportive of democracy and liberal values, more in favor of unconstrained strong leaders, and particularly concerned with immigration, especially with the cultural threat posed by it. These results are robust to controlling for the initial average attitudes of each region, computed from the oldest available survey for each country. Keywords: Globalization; Attitudes; Authoritarianism. We thank Francesco Bellogi for excellent research assistance. We thank Martin Elff, Jeffry Frieden, Peter Hall, Dani Rodrik, John Sides, Chris Way, as well as seminar participants at Harvard University, the MPSA 2018 Meeting in Chicago, and the EPSA 2018 meeting in Vienna for helpful comments. The usual disclaimer applies Assistant Professor of Economics. Department of Social and Political Sciences, Bocconi University. italo.colantone@unibocconi.it Assistant Professor of Political Science. Department of Social and Political Sciences, Bocconi University. Corresponding author. piero.stanig@unibocconi.it 1

2 What is behind the anti-establishment backlash currently observed across Western democracies? A lively academic debate is providing different answers to this question. In particular, several studies have shown that the success of nationalist and radical-right parties and candidates is empirically linked to economic distress, especially as driven by globalization (Autor et al. 2016; Colantone and Stanig 2018a, 2018b; Dippel et al. 2015; Frieden 2018; Malgouyres 2014; Rodrik 2018). The main argument behind these studies is that globalization as captured by the China shock generates economic grievances that in turn lead to anti-establishment voting. Indeed, while trade is overall welfare-enhancing at the level of countries, any episode of trade liberalization determines winners and losers, who tend to be concentrated in specific social groups and geographic areas. Failure to manage the adjustment costs of trade, through effective compensation and redistribution policies, would then drive voters dissatisfaction with the incumbent elites. Yet, what remains unclear in this literature is what are the channels through which the globalization shock translates into the observed voting behavior, and why voters have been pushed in a radical-right direction rather than a leftist one (Betz and Meret 2012; Colantone and Stanig 2018a). This is indeed somewhat puzzling, especially considering that insufficient redistribution is a key element of the argument that singles out globalization as a fundamental driver of the backlash. In this paper, we set out to answer these questions. We investigate the impact of globalization on people s attitudes across fifteen Western European democracies, over the time-span We employ individual-level data from the European Values Study (EVS) and the European Social Survey (ESS). We regress attitudes over exposure to the China shock, which we measure at the regionyear level as in Autor et al. (2013) and Colantone and Stanig (2018a). We attribute to each respondent the import shock of the region of residence, measured over the years preceding the survey. We focus on three facets of public opinion: (1) meta-political at- 2

3 titudes, i.e., support for democracy and for some of the foundational values underlying liberal democracy; (2) more private attitudes identified by public opinion scholars and political psychologists as characteristic of an authoritarian orientation; and (3) attitudes about immigration. We find that individuals living in areas more exposed to the shock are more supportive of unconstrained strong leaders and less supportive of democracy and liberal values. Moreover, they are more concerned with immigration, which is perceived even more as a cultural than an economic threat. We investigate the heterogeneity in responses by different groups of people, defined in terms of educational level and job status. We find that the import shock tilts different types of individuals in different ways, with less educated individuals often showing the strongest responses to the shock. Our results are robust to instrumenting Chinese imports to Europe using Chinese imports to the US, as originally proposed by Autor et al. (2013), and they are unaffected by the inclusion of controls for the historical average attitudes in each region. Importantly, our findings do not seem to be purely reflecting a general authoritarian shift. Indeed, we do not find any effects of the China shock on more private attitudes such as those concerning child-rearing and on other dimensions of authoritarianism, such as attitudes concerning the punitiveness of criminal justice. We make three contributions. First, we shed light on the link between economic shocks and voting behavior, as mediated by changes in attitudes. Our results help explain why the China shock has pushed voters towards nationalist and radical-right parties rather than pro-redistribution parties of the left. In particular, we show that the economic grievances driven by trade integration tilt people s attitudes in an authoritarian and nativist direction, which pushes them naturally towards extreme-right positions, and away from left parties with a reputation as supporters of solidarity and multiculturalism (Kriesi et al. 2012). 3

4 Second, we contribute to the ongoing debate on the economic vs. cultural roots of the political realignment observed in advanced Western economies: what is often referred to as a populist Zeitgeist (Gidron and Bonikowski 2016; Mudde 2004). The argument we support is that cultural and economic explanations should be seen as tightly related rather than alternative to each other. Indeed, our results suggests that cultural variables are themselves affected by economic distress. In this respect, we believe our paper provides evidence on what can be called the economic roots of the cultural backlash. To clarify, the causal chain we have in mind can be summarized as follows. There is a structural transformation in the economy induced exogenously by the emergence of China as a leading manufacturing exporter. This global phenomenon induces asymmetric adjustment costs across regions of advanced economies, thereby causing economic grievances that are unevenly distributed within each country, depending on the ex-ante industrial specialization of regions. Faced with these economic and social changes, people display a reaction that takes the form of more authoritarian and nativist attitudes. In turn, this shift in attitudes breeds support for nationalist and radical-right parties. Finally, our third contribution is more methodological. In the debate about cultural vs. economic drivers of the recent political changes, several studies are adopting a research design that involves regressing vote choices against broad sets of explanatory variables. These include jointly both cultural attitudes and measures of economic distress. Lack of significance of the economic indicators in these regressions is then taken as evidence that economic factors do not matter for vote choice. The article by Mutz (2018) is probably the most prominent example of this approach. An important implication of our results is that attitudes should be considered bad controls (to use the terminology by Angrist and Pischke, 2008) in regressions aimed at investigating the impact of economic conditions on voting. In fact, changes in attitudes are themselves an important channel through which economic variables might affect voting. As a result, 4

5 lack of significant correlation between economic conditions and vote choice in this type of analysis cannot be taken as evidence that economic conditions are not relevant, since post-treatment variables are included in the model. 1 Our point on this issue is in line with a recent critique by Morgan (2018). Clearly, we are not claiming that cultural reactions are only driven by economic determinants: there can be other factors affecting attitudes that are orthogonal to economic conditions. Yet, even a partial dependence of cultural variables on economic factors may invalidate any strong conclusions taken from regressions that condition on both at the same time. Theoretical framework and research design Recent political developments have sparked a debate about cultural versus economic explanations of voting behavior. To an extent, this debate echoes older ones (e.g., Lijphart 1979; Inglehart and Rabier 1986; Evans 2000). On one side, proponents of the materialist view suggest that party choice is largely driven by material interest and class position, and that changes in party competition are driven by changes in labor markets, income distribution, and economic opportunities. On the other side, various authors have claimed that changes in the political arena are the manifestation of a cultural backlash, for instance racial resentment in the US, or hostility to demographic changes driven by immigration in Western Europe. In this second perspective, concerns with group status or perceived cultural threats are largely responsible for the success of radical-right parties, extremist candidates, and in general for anti-establishment vote choice. Evidence has been provided in support of both views. On the materialist side, the surge of nationalist, isolationist, and radical-right parties is empirically linked to economic distress (Burgoon et al. 2018; Guiso et al. 2018; Hobolt and Tilley 2016), especially 1 See Angrist and Pischke (2008) and Samii (2016) for a discussion of post-treatment bias and bad controls. 5

6 as driven by trade globalization (Autor et al. 2016; Colantone and Stanig 2018a, 2018b; Dippel et al. 2015; Frieden 2018; Malgouyres 2014; Rodrik 2018). Yet, the cultural dimension also seems to play an important role in the electoral arena (Inglehart and Norris 2017; Mutz 2018). Some contributions have attempted to reconcile the two perspectives in different ways. In his analysis of the politics of the globalization backlash, Frieden (2018) notices that there is a failure of compensation of the losers from the economic side, but also a failure of representation from a political and cultural perspective. Margalit (2012) suggests that globalization is a multi-dimensional phenomenon involving not only international trade, but also migration and cultural consumption. As such, it might lead to a backlash that is itself multi-dimensional, encompassing both economic and cultural components. Gidron and Hall (2017, 2018) conceptualize more broadly how cultural and economic factors interact in determining people s anxiety about their social status, which in turn is found to be a proximate cause of nativism and radical-right support. This approach differs from the one adopted by some proponents of the cultural explanation, who interpret status threat purely as a cultural issue (e.g., Mutz 2018). In the same vein as Gidron and Hall (2017), Franzese (2018, 13) has noticed that the question is ill-formed: it s not status threat or economic hardship, it s and, or even because. In line with these contributions, we believe that economic distress and cultural reactions such as concerns with immigration can and should be considered part of the same process. As a matter of fact, this is also hinted by strong proponents of the cultural backlash approach. For instance, Inglehart and Norris (2016 3) suggest that the analytical distinction drawn between economic inequality and cultural backlash theories may also be somewhat artificial (...) if structural changes in the workforce and social trends in globalized markets heighten economic insecurity, and if this, in turn, stimulates a negative backlash among traditionalists towards cultural shifts. 6

7 In this paper, we contribute to this debate by investigating the effects of the China shock on people s attitudes. We show that cultural-backlash related variables are themselves affected by economic distress. In other words, demands for cultural protectionism not to mention appeals to ethnic or racial superiority cannot be interpreted at face value as consequences of a concern for culture however defined. Rather, they seem to be, at least partly, the cultural manifestation of grievances that are driven by the economic distress induced by globalization. Several contributions have studied the relationship between economic concerns, authoritarian and nativist attitudes, and political behavior. Some studies have investigated how anti-immigrant sentiments drive support for the radical right, and have tried to adjudicate between the relative roles of cultural threats and labor market competition as determinants of such sentiments, with somewhat mixed results (e.g., Golder 2003; Hainmueller and Hiscox 2007; Lucassen and Lubbers 2012; Malhotra et al. 2013; Rydgren 2008). More recently, Guiso et al. (2017), using ESS data, have shown that individuallevel economic insecurity is strongly related not only to opposition to immigration, but also to distrust of political parties and legislatures. In their paper, economic insecurity is measured through a composite index that combines self-reported economic distress with more objective individual characteristics, like sector of employment and skill level. Along the same lines, De Vries et al. (2018) find, in data from the US and European countries, that individual experience with economic hardship is associated with more authoritarian stances. In extant work, economic distress is predominantly measured using self-reported indicators. This may be problematic for several reasons, most importantly because economic perceptions could be endogenous with respect to political affiliations (Bartels 2002, Bisgaard 2015, Stanig 2013). We improve upon this literature by focusing on an objective measure of economic conditions at the regional level: the exposure to the China 7

8 shock. This depends on the ex-ante composition of employment in each region, which is kept fixed over time, and on the growth of Chinese exports to the world, which is mostly driven by internal factors of change in China (Autor et al. 2013). The surge of China as a global exporter with the consequent decline of manufacturing in exposed regions of advanced economies allows us to investigate the causal link between an exogenous structural change in economic conditions and people s attitudes. We find that individuals living in regions more exposed to the shock are less supportive of democracy and liberal values, more in favor of unconstrained strong leaders, and particularly concerned with immigration, especially due to the perceived threat it poses to national culture. Our results resonate with studies from psychology and political science showing that a perception of vulnerability, like the one induced by economic hardship, can induce the activation of an authoritarian syndrome (Ballard-Rosa et al. 2017; Hetherington and Suhay 2011; Napier and Jost 2008). Importantly, one element of the authoritarian personality syndrome is nativism and out-group hostility. Yet, at the same time, we do not detect any robust effects of the China shock on more private attitudes, such as those related to child-rearing, and on other dimensions of authoritarianism, including attitudes about the punitiveness of criminal justice. Overall, our evidence hints to a reaction that is genuinely political, and mostly directed towards specific issues such as immigration and the desirability of liberal democracy. Our paper is most closely related to Colantone and Stanig (2018b) and Cerrato et al. (2018). Colantone and Stanig (2018b), using data from the British Election Survey (BES), find that globalization, besides having an impact on voting, also affects attitudes and perceptions about immigration. In particular, the China shock received by the region of residence of an individual is a strong predictor of: opposition to immigration, perceptions of immigration as a threat to the national culture and the economy, and inflated perceptions of the immigrants arrival rate. These attitudes and perceptions, at 8

9 the same time, are not clearly related to the actual presence or arrival rate of immigrants. Along the same lines, Cerrato et al. (2018) find that the China shock increases hostility towards immigrants and minorities especially Hispanics and Asians in American National Election Studies data. We move forward with respect to these studies by using cross-country data, and, most importantly, by offering a broader assessment of multiple sets of attitudes beyond immigration. This allows us to characterize the effects of globalization on public opinion along several dimensions, which show different responses to the same shock. The China shock Our main explanatory variable is exposure to the China shock. Following Autor et al. (2013), we measure it at the region-year level through this indicator: Import Shock crt = j L rj(pre sample) L r(pre sample) IMPChina cjt L cj(pre sample), (1) where c indexes countries, r regions, j industries, and t years. IMPChina cjt is the change in (real) imports from China over the past n years, in country c and manufacturing industry j. This is normalized by the pre-sample number of workers in the same country and industry (L cj(pre sample) ). To retrieve the regional shock, we compute a weighted summation of all the industry-level changes in imports. The weights capture the relative importance of each manufacturing industry out of total employment in each region in the initial year (L rj(pre sample) / L r(pre sample) ). This index is based on a theoretical model developed by Autor et al. (2013). It captures the displacement generated by Chinese imports on the supply side of importing countries. Intuitively, stronger import shocks are assigned to regions that were specialized ex-ante in industries for which Chinese imports have increased relatively more 9

10 over time. The measure of industry specialization of each region is kept fixed at its pre-sample value, to avoid endogeneity issues. For a given specialization, the shock is stronger in years in which imports from China increase faster. In a series of papers, Autor et al. (2013, 2014, 2016, 2018) have employed this measurement approach to shed light on several dimensions of the economic adjustment costs induced by the China shock in the US. In particular, they have shown that a stronger exposure to the shock leads to lower employment rates, earnings losses, higher disability transfers, lower marriage and fertility rates, and even higher mortality at the regional level. Consistent evidence on the role of Chinese imports has been found across European countries, for instance by Bloom et al. (2016) and Utar (2018). Recently, a number of papers have adopted the methodology by Autor et al. (2013) to study the political implications of the China shock. Autor et al. (2016) have uncovered significant effects in terms of growing polarization in the US, and have provided evidence on the role of Chinese imports for the election of Donald Trump in Colantone and Stanig (2018a) have found that stronger exposure to the import shock has led to higher support for nationalist, isolationist, and radical-right parties across Western European countries. Similar evidence on radical-right parties has been obtained by Malgouyres (2014) and Dippel et al. (2015) on France and Germany, respectively. Colantone and Stanig (2018b) have found that the China shock has also played an important role for the success of the Leave option in the Brexit referendum of The intuition behind this body of evidence is that the import shock generates economic grievances that push voters against the establishment. In this paper, we move one step forward by studying the link between the import shock and individual attitudes. We regard this as a crucial contribution, since changes in attitudes may help understand why the anti-establishment backlash has taken the specific political form documented by the literature. 10

11 Our analysis covers 15 industrialized countries of Western Europe (list in Table A1 of the Online Appendix). We measure the China shock at the NUTS-2 level of regional disaggregation. 2 Overall, our sample spans 143 regions. We source import data from Eurostat Comext and the CEPII-BACI database. To compute the employment shares at the regional level, we rely on data from Eurostat and a number of national sources. Full details are provided in Table A1. We work at the NACE Rev 1.1 sub-section level of industry disaggregation, which cuts the manufacturing sector in 14 industries. In the baseline analysis, we measure the growth in imports from China over the past two years before each survey. Each respondent is assigned a value of the import shock based on the region of residence. The resulting import shock variable, computed as in Equation (1), has a mean of across EVS respondents in all waves. This corresponds to 143 real euros per worker, with a standard deviation of In the ESS sample, the mean is 0.177, with the same standard deviation. To deal with the possible endogeneity of Chinese imports, we follow the same approach as in Autor et al. (2013) and Colantone and Stanig (2018a, 2018b). Specifically, we employ the following instrumental variable: Instrument for Shock crt = j L rj(pre sample) L r(pre sample) IMPChinaUSA jt L cj(pre sample). (2) The difference with respect to Equation (1) is in the numerator of the second term, where we consider imports from China to the US instead of each European country. The aim is to capture the variation in imports from China to Europe that is driven by exogenous changes in supply conditions in China, rather than by domestic factors specific to each European country, which could be correlated with individual attitudes. Data on US imports are sourced from the Center for International Data of UC Davis. 2 The only exceptions are France, Germany, and the UK, for which either attitudes data or employment shares data are only available at the NUTS-1 level. 11

12 Data and models Individual-level data are sourced from the European Social Survey (ESS) and the European Values Study (EVS). Specifically, we use the first four waves of the ESS, which cover the years from 2002 to For EVS, we employ the cumulative trend file, which spans the period From the latter collection, in the econometric analysis we only use data starting from the 1990s, which can be matched to import shock data at the regional level. Importantly, though, we use earlier surveys to calculate the initial (regional-level) summaries of attitudes that we include as controls. Each respondent is assigned a value of the import shock based on the survey year and the region of residence, at the the NUTS-2 level. 3 The regressions have the following general form: Atttitude icrt = α ct + β 1 Import Shock cr(i)t + X it γ + ε icrt, (3) where i indexes individuals, c countries, r regions, t years, and ε icrt is an error term. Depending on the specific analysis, the outcome variable Attitude icrt is one of those described in the following section. The function r() maps each individual (i) to her NUTS-2 region of residence (r). Import Shock cr(i)t is the growth in Chinese imports at the regional level over the two years prior to the survey. α ct are country-year fixed effects, which are equivalent to survey fixed effects. These control for any factors that affect symmetrically all the districts within a country at the time of a given survey. For instance: the country-level overall economic performance, the orientation of the incumbent national government, and the general political climate within each country. The inclusion of country-year fixed effects implies that we identify the impact of the import shock only out of variation across regions within the same country and year. Moreover, 3 In some cases, information on the region is only available at the NUTS-1 level, and the import shock is computed accordingly. See also footnote 2. 12

13 standard errors are clustered at the region-year level, to account for possible correlation across respondents residing within the same region. Finally, X it is a vector of individual-level, pre-treatment controls: age, a dummy for females, and dummies indicating different levels of educational attainment, as classified by ISCED in the case of the ESS, and as a coarser four-category classification in the case of the EVS. A concern one might have with our analysis, due to the way we measure the China shock, is that there are stable differences across regions which are correlated with the time-invariant part of the import shock measure, that is, the sectoral composition of the economy of a given region in the late 1980s or early 1990s. In order to address this concern, we also estimate models with the following form: Attitude icrt = α 0ct + δz cr(i)0 + β 1 Import Shock cr(i)t + X it γ + ε icrt, (4) where Z cr(i)0 is the regional-level average of the outcome variable at time t = 0, namely in the oldest survey for which it is available in each collection. In this way, we can account for possible persistent patterns in attitudes across regions, which in turn might be correlated with the strength of the China shock, since this is driven, at least partly, by the initial composition of the regional economy. The exact survey from which the initial average is taken varies by item in the case of the EVS. For the ESS we always calculate initial values in Consistently, when the initial value of the responses is included in the estimation, we drop observations from the 2002 wave of the ESS. Measuring attitudes We focus on three dimensions of public opinion: (1) meta-political attitudes and liberal values; (2) private authoritarian attitudes; and (3) immigration attitudes. In what 13

14 follows, we describe each set of outcome variables. The first two variables we consider are support for democracy and support for unconstrained strong leaders, both sourced from the EVS. These variables capture, respectively, individual opinions about the desirability of democracy as a form of government, and preferences regarding a strong leader free from the control of parliament and elections. We code both variables so that lower values denote more authoritarian attitudes, on a four-point scale. We also compute an overall Democratic index as the sum of the two items, to capture the general support for the democratic system. Next, we investigate support for a set of values that are arguably central for the legitimacy and functioning of a liberal democracy. These values refer to the importance of: (1) equality and equal opportunities; (2) understanding different people; (3) being free; and (4) following rules. We regard these attitudes as underpinning the foundations of liberal democracy. Data on each item are available in the ESS, on a six-point scale. Higher values denote less importance attached to a given value. We also create an overall index of Liberal Values, as the first principal component of the four individual items. The single items are all positively correlated, with the importance of following rules showing the weakest correlation with the others. Our second set of attitudes addresses the concept of authoritarianism as a personality trait, as discussed in political psychology. We start by focusing on child-rearing. In particular, we employ four EVS items on the importance assigned to the following qualities of children: manners, imagination, obedience, and independence. These items are often used as proxies for an authoritarian personality (Feldman and Stenner 1997). They are coded as binary variables equal to 1 when a given quality is considered to be especially important. Again, we use the first principal component of the four items to create an index of authoritarian preferences regarding child-rearing, which we name Children Qualities. Intuitively, assigning importance to imagination and independence is nega- 14

15 tively correlated with assigning it to manners and obedience. The index is coded in such a way that higher values indicate a more authoritarian stance. Besides child-rearing, we also consider five additional items aimed at capturing private authoritarian (vs. libertarian) attitudes. The first of these items taps opinion about abortion, which tends to be a central marker of traditionalism and conservatism in Western countries (e.g., Fiorina and Abrams 2008; Engeli et al. 2012). Specifically, we employ a measure available in the EVS on a ten-point scale, from never justifiable to always justifiable, with higher values indicating a more permissive stance. Then, we employ three ESS items capturing the importance assigned to: following traditions, being creative, and living in safe surroundings. These variables are measured on a six-point scale, with higher values denoting lower importance. Finally, we consider preferences regarding harshness of punishment in criminal matters, as measured in the ESS on a five-point scale. Higher values denote less punitive attitudes. In the third section of empirical results, we consider attitudes about immigration. In particular, we consider two items from the ESS: (1) the opinion on whether the country s cultural life is undermined or enriched by immigrants (Immigration Culture); and (2) the opinion on whether immigration is bad or good for the country s economy (Immigration Economy). Both variables are measured on a 10-point scale, and coded in such a way that higher values denote more positive views of immigration. Results We start by presenting the results on meta-political attitudes. Specifically, Tables 1 and 2 report, respectively, OLS and 2SLS estimates for three outcome variables: (1) support for democracy; (2) support for an unconstrained strong leader; and (3) a Democratic index computed as the sum of (1) and (2). Lower values of each variable refer to more 15

16 authoritarian and less democratic attitudes. For each outcome, we report results from two different specifications, both in OLS and 2SLS. The first specification includes only the standard controls for age, gender, and education, plus the country-year fixed effects, as in Equation 3. In the second specification, we add the initial average attitudes in each region, as in Equation 4. The coefficient of the import shock is precisely estimated across the board. The negative sign indicates that stronger import shocks lead to less democratic and more authoritarian attitudes. Compared to the OLS estimates of Table 1, the IV coefficients in Table 2 are systematically larger in absolute value. This suggests that there are unobserved factors, such as positive demand shocks, that correlate at the same time with higher imports from China and with pro-democracy attitudes. The first-stage coefficient on our instrument is always positive and significant, and the F-statistic does not signal a weakness problem, in line with earlier studies (e.g. Autor et al. 2013; Colantone and Stanig 2018a; 2018b). Importantly, the inclusion of initial attitudes does not significantly affect the results, neither in the OLS estimations nor in the IV ones. Table 1: Political authoritarianism - OLS (1) (2) (3) (4) (5) (6) Dep var.: Strong Leader Democracy Democratic Import Shock * ** *** *** *** *** [0.044] [0.046] [0.019] [0.019] [0.057] [0.060] Female 0.020** *** *** * [0.010] [0.010] [0.007] [0.007] [0.014] [0.014] Age ** * 0.002*** 0.003*** 0.002*** 0.002*** [0.000] [0.000] [0.000] [0.000] [0.001] [0.001] Estimator OLS OLS OLS OLS OLS OLS Education Dummies yes yes yes yes yes yes Country-Year Effects yes yes yes yes yes yes Initial Attitudes - yes - yes - yes Obs. 46,867 43,639 47,104 44,154 44,805 41,669 R Standard errors are clustered by region-year. *** p<0.01, ** p<0.05, * p<

17 Table 2: Political authoritarianism - IV (1) (2) (3) (4) (5) (6) Dep var.: Strong Leader Democracy Democratic Import Shock ** *** *** *** *** *** [0.115] [0.112] [0.067] [0.062] [0.164] [0.152] Female 0.019** *** *** * [0.009] [0.010] [0.007] [0.007] [0.014] [0.014] Age ** * 0.002*** 0.003*** 0.002*** 0.002*** [0.000] [0.000] [0.000] [0.000] [0.001] [0.001] Estimator 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS Education Dummies yes yes yes yes yes yes Country-Year Effects yes yes yes yes yes yes Initial Attitudes - yes - yes - yes Obs. 46,867 43,639 47,104 44,154 44,805 41,669 R First-stage results US imports from China 0.093*** 0.114*** 0.092*** 0.114*** 0.092*** 0.113*** [0.021] [0.023] [0.021] [0.023] [0.021] [0.023] Kleibergen-Paap F-Statistic Standard errors are clustered by region-year. *** p<0.01, ** p<0.05, * p<0.10 In Table 3, the dependent variable is Liberal Values. This is an index capturing overall support for the foundational values of liberal democracy, computed as explained in the previous section. The table reports both OLS and 2SLS estimates, with and without controls for initial attitudes in the region. The coefficient on the import shock is always positive and significant, pointing to less support for liberal values in areas that receive stronger import shocks. Also in this case, the magnitude of the coefficient is larger in the IV models than in the OLS models. The drop in the number of observations when controlling for initial attitudes is due to the exclusion of individual data from the 2002 ESS survey, which we use to compute the initial averages within each region. In Table A2 of the Online Appendix, we report separate estimates for each of the four items that enter the Liberal Values index. The strongest relationship we detect is between the import shock and the importance of equality, but the relationship is positive (albeit not precisely estimated) also for the other three items. 17

18 To sum up, the message that can be drawn from this first set of results is that the China shock is associated with and plausibly causally affects meta-political attitudes and support for liberal values. In particular, respondents residing in regions that were more exposed to the import shock tend to be more sympathetic to the idea of an unconstrained strong leader, and less unequivocally supportive of democracy and liberal values than otherwise similar individuals residing in areas that were less exposed to the shock. Importantly, the identified effects are robust to the inclusion of controls for the initial average attitudes in each region. This reassures us that our findings are not simply driven by a persistent authoritarian outlook in areas where manufacturing has historically played a larger economic role. In other words, our results do not seem to be driven by a sort of blue-collar culture effect that we mistake for a globalization effect. In this respect, it is also important to notice that the China shock does not hit in the same way all manufacturing regions, but varies with their historical industry specialization. For instance, Chinese import pressure is not so relevant in automotive or machine tools industries, which might be, stereotypically, the industries where a blue-collar authoritarian culture would be expected to flourish (e.g., Lipset 1959). Authoritarian traits Table 4 reports a set of regressions using as outcome variable Children Qualities: the index that combines four EVS items regarding child-rearing. Higher values indicate a more authoritarian stance. We estimate two different specifications, according to Equations 3 and 4, both in OLS and in 2SLS. Regardless of the estimation method and the model, we do not detect any significant effect of the import shock on attitudes concerning the education of children. This result is confirmed in Table A3 of the Online Appendix, where we consider separately each of the items entering the index, without finding any strong 18

19 Table 3: Liberal values (1) (2) (3) (4) Dep var.: Liberal Values Import Shock 0.082** 0.079** 0.189** 0.162** [0.034] [0.035] [0.085] [0.080] Female *** *** *** *** [0.012] [0.013] [0.012] [0.013] Age ** *** ** *** [0.000] [0.000] [0.000] [0.000] Estimator OLS OLS 2SLS 2SLS Education Dummies yes yes yes yes Country-Year Effects yes yes yes yes Initial Attitudes - yes - yes Obs. 102,228 74, ,228 74,091 R First-stage results US imports from China *** 0.091*** - - [0.017] [0.017] Kleibergen-Paap F-Statistic Standard errors are clustered by region-year. *** p<0.01, ** p<0.05, * p<0.10 evidence of significant effects of import exposure. In Table 5, we report the estimates of models where we investigate the impact of the import shock on five items reflecting additional dimensions of the overall authoritarian orientation of respondents. In none of the estimations there is any detectable relationship between the import shock and attitudes. Overall, the findings presented in this section suggest that the China shock does not lead to more authoritarianism in terms of private attitudes, unlike what we found in terms of meta-political attitudes and liberal values. This evidence has two main implications. The first one is that the significant results on the political dimension are not driven by some stable and comprehensive differences in life orientations across areas more or less exposed to globalization. Indeed if it were simply that there are two stable types of regions, authoritarian ones and libertarian ones, it would be plausible to expect them to differ systematically on a whole host of attitudes, 19

20 including those on child-rearing, abortion, and the like. On the opposite, what emerges from our analysis is that economic distress leads to lower support for democracy and liberal values, but has no bearing on individual private attitudes, such as those about desirable characteristics of children. Second, our results suggest that the detected effects of the import shock might not be the consequence of a general shift in the direction of the authoritarian personality, as per the conventional wisdom in political psychology. Indeed, it seems that the import shock is associated with and might plausibly cause disaffection with the extant political system and its underlying egalitarian civic values. Yet, there is no detectable association with other attitudes that the literature suggests are part of an authoritarian personality syndrome. Skepticism regarding democracy and liberal values, and a desire for strong unconstrained leadership, can be considered genuine political manifestations of a discontent driven by economic distress. Table 4: Children qualities (1) (2) (3) (4) Dep var.: Children Qualities Import Shock [0.025] [0.023] [0.053] [0.052] Female ** * *** *** [0.006] [0.006] [0.006] [0.007] Age 0.005*** 0.005*** 0.003*** 0.004*** [0.000] [0.000] [0.000] [0.000] Estimator OLS OLS 2SLS 2SLS Education Dummies yes yes yes yes Country-Year Effects yes yes yes yes Initial Attitudes - yes - yes Obs. 55,746 47,341 38,986 31,831 R First-stage results US imports from China *** 0.085*** - - [0.022] [0.023] Kleibergen-Paap F-Statistic Standard errors are clustered by region-year. *** p<0.01, ** p<0.05, * p<

21 Table 5: Authoritariatism - other dimensions (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Dep var.: Abortion Traditions Creativity Safety Punishment Import Shock [0.142] [0.252] [0.040] [0.066] [0.025] [0.054] [0.044] [0.068] [0.129] [0.155] Female *** ** *** *** 0.076*** 0.076*** *** *** *** *** [0.026] [0.030] [0.012] [0.012] [0.011] [0.011] [0.012] [0.012] [0.019] [0.019] Age *** *** *** *** 0.008*** 0.008*** *** *** *** *** [0.001] [0.001] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.001] [0.001] Estimator OLS 2SLS OLS 2SLS OLS 2SLS OLS 2SLS OLS 2SLS Education Dummies yes yes yes yes yes yes yes yes yes yes Country-Year Effects yes yes yes yes yes yes yes yes yes yes Obs. 65,656 49, , , , , , ,564 24,462 24,462 R First-stage results US imports from China *** *** *** 0.090*** *** - [0.021] - [0.016] - [0.016] [0.016] - [0.027] Kleibergen-Paap F-Statistic Standard errors are clustered by region-year. *** p<0.01, ** p<0.05, * p<

22 Immigration attitudes Table 6 reports estimates concerning immigration attitudes. We focus on two outcome variables: (1) the perceived cultural impact of immigrants (Immigration Culture); and (2) the perceived economic impact of immigrants (Immigration Economy). Both items are sourced from the ESS, and coded in such a way that higher values reflect more positive views of immigration, on a 10-point scale. The first four columns of Table 6 report the results for the cultural impact of immigration, both in OLS and in 2SLS, with and without controls for the initial attitudes in the region. Again, the drop in the number of observations when controlling for initial conditions is due to the exclusion of data from the ESS survey of 2002, which is used to compute the initial averages. The estimated coefficient on the import shock is always negative and significant. This points to a significant effect of the China shock on perceptions of immigrants as a threat to national culture. All else equal, respondents who reside in regions that received stronger import shocks are more concerned with the cultural threat posed by immigrants, as compared to similar individuals living in regions less exposed to import competition. Instead, when it comes to the economic impact variable, in columns 5-8, the coefficient on the import shock is never statistically significant. This points to the fact that, in general, there is not a very strong association between the economic distress captured by the import shock and the belief that immigration is harmful for the national economy. In the next section, we provide additional evidence on this finding by exploring possible heterogeneous effects across different categories of respondents. 22

23 Table 6: Immigration attitudes (1) (2) (3) (4) (5) (6) (7) (8) Dep var.: Immigration Culture Immigration Economy Import Shock *** ** *** * [0.081] [0.070] [0.164] [0.157] [0.079] [0.079] [0.137] [0.134] Female *** *** *** *** [0.025] [0.029] [0.025] [0.029] [0.023] [0.028] [0.023] [0.028] Age *** *** *** *** *** *** *** *** [0.001] [0.001] [0.001] [0.001] [0.001] [0.001] [0.001] [0.001] Estimator OLS OLS 2SLS 2SLS OLS OLS 2SLS 2SLS Education Dummies yes yes yes yes yes yes yes yes Country-Year Effects yes yes yes yes yes yes yes yes Obs. 106,548 78, ,548 78, ,213 78, ,213 78,026 R First-stage results US imports from China *** 0.089*** *** 0.089*** - - [0.016] [0.016] - - [0.016] [0.017] Kleibergen-Paap F-Statistic Standard errors are clustered by region-year. *** p<0.01, ** p<0.05, * p<

24 Heterogeneity The models presented thus far are estimating a pooled effect of the import shock on all the respondents who reside in a given region. Yet, different categories of individuals might display different reactions to the same shock. In this section, we explore potential heterogeneous effects according to education level and job status. We focus on the four main outcome variables of our analysis: (1) the Democratic index; (2) the Liberal Values index; (3) the cultural impact of immigrants (Immigration Culture); and (4) the economic impact of immigrants (Immigration Economy). In Table 7 we explore heterogeneity by level of education. We consider four educational categories: lower education (up to middle school); middle education (high school); higher education (university level); and a residual category called other education, including missing responses and values that cannot be standardized according to the ISCED classification. We allow the effect of the China shock to vary by respondent type. To this purpose, we interact the import shock with the four dummies denoting each education category, and we include the four interactions in the specification by omitting the linear term of the import shock. Essentially, we are estimating one slope per education group. We also include the linear terms of the education dummies excluding the low education one, so that all the dummy coefficients shall be interpreted as differences with respect to the low education category. When we consider overall support for democracy as an outcome, in column 1 of Table 7, the coefficient on the import shock is negative and clearly bounded away from zero for the two lower categories of education. 4 Moreover, the estimates on the education dummies suggest that there is a clear gradient on education level. Specifically, for any given level of the import shock, more educated respondents are more supportive 4 We also re-estimate this model using a binary indicator for respondents with a university education vs. everyone else. The results, available upon request, are analogous to the reported ones. 24

25 of democracy. This finding is in line with earlier literature. It is worth noticing that respondents in the residual category appear to be closer to the low education group than others. A similar pattern obtains in column 2, where the outcome variable is the index of liberalism: the effect of the import shock is always in an illiberal direction, but it is not statistically distinguishable from zero for university-educated individuals (and for the residual category). As usual with heterogeneity analysis, since the features of the unit are not randomly assigned, we cannot attribute causal interpretation to the heterogeneity itself. Yet, these results reveal that less educated individuals might be more susceptible to anti-democratic and illiberal appeals. The pattern of findings is somewhat different in the case of immigration attitudes. In column 3, we consider the cultural impact of immigrants. The import shock is significantly associated with immigration concerns across the whole spectrum of education categories, but in this case the effect is stronger for higher-education individuals. The gradient on education levels is consistent with what the literature has abundantly documented: at zero import shock, higher educated individuals are less concerned with immigration. Yet, the import shock has a depolarizing effect in a nativist direction. Specifically, in settings with sufficiently strong import shocks, respondents with higher education are more similar to less educated individuals in terms of cultural concerns with immigration. Finally, in the fourth column we explore heterogeneity in immigration concerns from the economic perspective. In the previous section, we show that there is no clear evidence that the import shock increases concerns with the economic consequences of immigration. Yet, when we disaggregate the effect by education level, we find that there is a detectable (albeit short of statistical significance) relationship between the import shock and the attitudes of high-education respondents. In addition, the effect is particularly strong for respondents whose education level could not be assigned to one of 25

26 the three categories. Again, this pattern of findings depicts a depolarizing effect of the import shock, by which a larger fraction of the population converges to the higher level of immigration concerns of less educated people. In Table 8 we address heterogeneity by job status. We focus on six categories of respondents: employed, retired, homemakers, students, unemployed, and a residual category containing, for instance, military personnel. 5 The outcome variables and the structure of the model are exactly the same as in Table 7. That is, we estimate a coefficient of the import shock for each category. The excluded dummy in this case is that of employed people, so all the other dummies shall be interpreted as differences with respect to them. The first column reports the results for the Democratic index. The import shock has a negative and statistically significant association with support for democracy for all the job-status categories, with the exception of students and respondents in the residual category, who are seemingly shielded from the shock. The fact that also retired people and homemakers are more skeptical of democracy in regions hit harder by the import shock suggests that the effects of globalization are not restricted to workers who are directly affected by the shock. Rather, global competition seems to have a broader impact on the population of affected regions, in a sociotropic fashion. Importantly, though, the unemployed appear to be not only the most skeptical of democracy at zero import shock as signaled by the magnitude and size of their dummy but also the most sensitive to the import shock itself, as can be inferred from their specific slope. The pattern we detect for the Democratic index is not replicated when we move to the Liberal Values index. In fact, the illiberal effect of the import shock obtained in the pooled regression is clearly detectable for the employed and for students, but it is not statistically different from zero for the other occupational categories. 5 The categories are identified starting from the variable mnactic in the ESS and X038 in the EVS. Some of the original categories are collapsed for consistency. 26